about memory zone usage

[yoma sophian]

hi all:
in buddy allocator heart function, __alloc_pages_nodemask, I found we
use zonelist-> _zonerefs to iterate all the zone in NUMA or UMA.

My questions are:
1. the advantage of travelling  zonelist-> _zonerefs instead of
iterating zones on each node is because it is faster, right?
2. zonelist-> _zonerefs has recorded information of each zone, under
what circumstances we will iterated zone of node to get zone's data
instead of  zonelist-> _zonerefs

appreciate your help in advance,